A soft prosthesis hand control method based on myoelectric signals

Abstract

The application discloses a soft prosthesis hand control method based on myoelectric signals, and the soft prosthesis hand is made of soft silicone and driven by air pressure. The method comprises the following steps: (1) collecting myoelectric signals of hand movement process by a myoelectric sensor and identifying movement results; (2) modeling mechanics of a soft driver to obtain a relationship between air pressure borne by the soft driver and object contact force; (3) constructing a non-singular terminal sliding mode function of a deviation between an expected contact force at the end of the soft driver and an actual contact force; (4) approximating a system disturbance and a modeling error function by using a RBF neural network, and obtaining a control rate according to Lyapunov stability discrimination; and (5) combining the control rate with the action identified in the step (1) to control the soft prosthesis hand to complete a grasping task. The control method solves the problem that a traditional PID control has poor control effect in a nonlinear control system of the soft prosthesis hand, reduces the problems of slow system response and serious chattering, and is a complete closed-loop control method for grasping objects by the soft prosthesis hand.

Description

TECHNICAL FIELD

[0001] The application relates to a soft prosthesis hand force control method based on an electromyographic signal, in particular to a soft prosthesis hand force control method, and belongs to the technical field of soft robot control. BACKGROUND

[0002] With the in-depth research of researchers on bioelectric signals, electromyographic signals have been widely used. The electromyographic signal is a weak electric potential generated by the movement of muscle cells in the process of muscle contraction in the human body. Each movement intention of a person corresponds to a unique electromyographic signal. The electromyographic signal of an amputee is collected by using an electromyographic signal acquisition device, and the electromyographic signal is filtered, amplified, feature extracted and action recognized to determine the mapping relationship between the action of the prosthesis hand and the electromyographic signal. The electromyographic signal result of the recognized corresponding action is taken as an action command, so that the electromyographic control of the prosthesis hand can be realized.

[0003] In order to help patients recover the ability to live and work, the prosthesis hand can replace the affected limb of the patient to a certain extent, and basically meet the work demand of daily life.

[0004] The prosthesis hand can be divided into rigid hard prosthesis hand and soft flexible prosthesis hand according to the material, and the rigid hard prosthesis hand is mainly used on the market at present. The rigid hard prosthesis hand is composed of rigid connecting rods, hinges, motors and other metal instruments, and the soft flexible prosthesis hand is made of soft materials such as flexible fluid and human silicone. Compared with the rigid hard prosthesis hand, the soft flexible prosthesis hand has the advantages of light weight, safety, good bionics and low cost. Because the body is a flexible soft structure, it is simple to manufacture, low in cost and light in weight, and is safe and comfortable for the patient himself and others. Secondly, because the shape of the human hand is soft outside and hard inside, the soft flexible prosthesis hand has good bionics. Therefore, the soft flexible prosthesis hand has good development prospect.

[0005] Although the soft flexible prosthesis hand has very good application prospect, the soft flexible prosthesis hand is still in the laboratory stage at present, and there are only some soft prosthesis hands with single function on the market, which are mostly passive in performing the grasping task, and have no biological signals such as electromyographic signals for control. The patent document with the application number “201810716880.0” discloses a “human-like dexterous electromyographic prosthesis hand control method based on gesture recognition”. The neural network algorithm is adopted to recognize the action of the collected human electromyographic signal, and the electromyographic control of the prosthesis hand is realized to a certain extent, but the control system has no controller, and the whole system is open-loop. Therefore, the precision of the prosthesis hand in grasping objects is not high by simply using the electromyographic signal control, and the grasping of objects is prone to failure.

[0006] The main reason for the slow development of the soft flexible prosthetic hand is its strong nonlinear characteristics, which makes it difficult to accurately model and control. Secondly, since the soft prosthetic hand is relatively small, there is no suitable sensor to integrate it, making it difficult to measure its position, force and driving source and other important data. Finally, since most soft prosthetic hands are driven by air pressure, the softness of the soft prosthetic hand and the high hysteresis of the air pressure drive will further cause its uncertainty and disturbance, which seriously affects the effect of accurately controlling the soft prosthetic hand. The patent application No. "201710378315.3" discloses a "pneumatic soft finger, soft finger control system and control method", which detects the sensing air cavity pressure to measure the finger bending curvature and contact force, and realizes closed-loop control of the finger contact force by using the PID algorithm. But the soft driver has strong nonlinearity, and the traditional PID algorithm is a linear controller, which often shows poor controller parameter setting, low control precision and cannot achieve ideal control effect in nonlinear control system.

[0007] In view of the above problems, the present application provides a soft prosthetic hand control method based on electromyographic signal, which establishes a mechanical model of the soft driver to obtain the relationship equation between the air pressure and the object contact force of the soft driver, establishes a non-singular terminal sliding mode controller according to the relationship equation, and approximates the modeling error and disturbance function of the control system through RBF neural network. SUMMARY

[0008] The present application aims to provide a soft prosthetic hand control method based on electromyographic signal. The present application needs to solve two problems: one is to model the pneumatic soft driver, obtain the function relationship between the air pressure and the object contact force of the soft driver, and propose a non-singular terminal sliding mode control algorithm with air pressure as input and soft driver contact force as output, to solve the problem of existing soft prosthetic hand controlled by PID or even without controller, so as to make the soft prosthetic hand control system state converge to zero in a limited time; the second is to use RBF neural network to approximate the unknown disturbance function in the modeling function of the soft driver, and then derive the control rate and adaptive rate of the control system according to Lyapunov stability criterion, to reduce the chattering problem of the control system and enhance the robustness of the soft prosthetic hand control system.

[0009] The purpose of the present application is achieved by the following technical solutions:

[0010] A soft prosthetic hand control method based on electromyographic signal, comprising the following steps:

[0011] (1) Collecting the electromyographic signal of the hand movement process by the electromyographic sensor, and identifying the corresponding hand movement result;

[0012] (2) the mechanical modeling of the soft driver, the nonlinear mechanical analysis of the single soft driver is carried out by adopting the Yeoh model, and the relationship between the air pressure and the object contact force borne by the soft driver is obtained according to the mechanical balance, and the Yeoh model is a constitutive model;

[0013] (3) the non-singular terminal sliding mode function of the deviation between the expected contact force and the actual contact force at the end of the soft driver is constructed according to the relationship between the air pressure and the object contact force borne by the soft driver obtained in step (2), so as to accelerate the convergence speed of the control system;

[0014] (4) for the unknown function of the control system disturbance and the modeling error, the unknown function of the control system disturbance and the modeling error is approximated by adopting the RBF neural network, so as to reduce the chattering problem of the control system, finally, the Lyapunov function of the deviation between the expected contact force and the actual contact force at the end of the soft driver is constructed, and the control rate and the adaptive rate when the derivative of the Lyapunov function is less than or equal to zero are calculated, thereby, the ideal weight of the neural network is adjusted according to the adaptive rate, and the soft driver is controlled according to the control rate;

[0015] (5) the control rate is combined with the motion control soft prosthesis hand identified in step (1) to complete the grasping task.

[0016] The object of the application can also be further realized by the following technical measures:

[0017] Further, step (2) specifically comprises:

[0018] Step (2.1): the Yeoh model is selected to perform nonlinear mechanical analysis on the soft driver, the constitutive relationship of the rubber material is established based on the stress-strain relationship, and then the strain energy density function and the deformation tensor invariant of the soft driver are:

[0019]

[0020] Wherein W is the strain energy density function, lambda1, lambda2 and lambda3 are the main elongation ratios of the length, width and height of the soft driver respectively, I1, I2 and I3 are the deformation tensor invariants of the length, width and height of the soft driver respectively.

[0021] The main elongation ratio lambda1 of the soft base body of the soft driver after inflation and bending is:

[0022]

[0023] Wherein l is the air cavity length of the soft driver, t3 is the bottom layer thickness of the soft driver, Is the bending angle.

[0024] In combination with the above formula:

[0025]

[0026] λ3=1

[0027]

[0028] Step (2.2): Based on the typical binomial parameter form of the Yeoh model, its strain energy density function model is as follows:

[0029] W = C1(I1-3) + C2(I1-3) 2

[0030] Where C1 and C2 are the slope and intercept of the stress fitting line of the silicone Yeoh model material, respectively, which are determined by material experiments and data fitting. Through experiments, C1 is taken as 0.1 and C2 as 0.005.

[0031] Step (2.3): The stress in the length direction of the software driver is:

[0032]

[0033] Where σ is the stress, the stress of the software driver is obtained from the above formula:

[0034]

[0035] Step (2.4): The bending moment of the software actuator is represented by 5 bending moments, namely the main moment M generated by the gas in the gas chamber. a The tensile torque M of the top layer, bottom layer, and sidewall materials t M b M c And the torque M generated by the interaction between the end of the software driver and the object. f :

[0036]

[0037]

[0038]

[0039]

[0040]

[0041] Where p is the air pressure on the air chamber unit of the software actuator, B is the width of the software actuator, t is the wall thickness of the software actuator, h is the height of the air chamber, and L is the height of the air chamber. f denoted as the length of the software actuator tip, f as the force generated by the interaction between the software actuator tip and the object, and x as the increment of the software actuator wall thickness.

[0042] According to the moment balance:

[0043] M a -2M c -M t -M b = M f

[0044] Step (2.5): Further, the soft actuator mechanical equation is described by the following equation:

[0045]

[0046] Where g(σ) is the resultant moment generated by the soft actuator top layer, bottom layer, and sidewall material stretching, f is the force generated by the soft actuator end interacting with the object, d(t) is the modeling error, external disturbance unknown function, L f is the length of the soft actuator end, and p is the air pressure acting on the soft air cavity.

[0047] Step (2.6): The above model is rewritten as follows:

[0048]

[0049] Where, define

[0050]

[0051] Further, the relationship equation between the air pressure acting on the soft actuator and the contact force of the object is defined as:

[0052]

[0053] Where is the second-order derivative of the force generated by the soft actuator end interacting with the object, is the first-order derivative of the force generated by the soft actuator end interacting with the object.

[0054] Further, step (3) specifically includes:

[0055] Step (3.1): Based on the above relationship equation between the air pressure acting on the soft actuator and the contact force of the object, the deviation function of the desired contact force and the actual contact force at the end of the soft actuator is constructed, assuming that the desired force of the soft actuator after inflation and bending is f d , the actual force is f, and the deviation function is defined as:

[0056] e = f d -f

[0057] Step (3.2): According to the deviation function, the sliding surface function of the non-singular terminal sliding mode control is selected as:

[0058]

[0059] wherein is positive, p, q are positive odd numbers, and Combining the above formula:

[0060]

[0061] wherein

[0062]

[0063]

[0064]

[0065] wherein is an unknown function to be approximated.

[0066] Further, step (4) specifically comprises:

[0067] Step (4.1): using RBF neural network to approximate the unknown disturbance function wherein

[0068]

[0069]

[0070]

[0071] wherein x is the network input, j is the jth node of the network hidden layer, c j is the kernel function center, is the square of the Gaussian kernel width, h(x) = [h j ] T is the network Gaussian kernel output, W * is the ideal weight of the network, is the estimated value of the ideal weight of the network, is the error of the ideal weight and the estimated value of the network, ε is the network approximation error, is the disturbance function approximated by the neural network.

[0072] Step (4.2): constructing the Lyapunov function of the control system

[0073]

[0074] wherein γ is a positive number,

[0075] Step (4.3): designing the control rate

[0076] where sgn() is a sign function, and η is a switching term gain.

[0077] Combining the above formula

[0078]

[0079] Step (4.4): In order to make the non-singular terminal sliding mode controller satisfy the Lyapunov stability condition, take The adaptive rate is:

[0080]

[0081] Combining the above formula, then

[0082]

[0083] According to the above formula, the controller satisfies the Lyapunov stability condition.

[0084] Further, step (5) specifically comprises:

[0085] Step (5.1): The corresponding hand motion result identified in step 1 is taken as a motion instruction, and the instruction is sent to the lower computer main control board. After receiving the instruction, the main control board controls the air pump and the electromagnetic valve to inflate the soft artificial hand to realize bending and approaching the object.

[0086] Step (5.2): When the membrane pressure sensor data at the end of the soft artificial hand is greater than the threshold value, it is considered that the soft artificial hand is grasping the object. At this time, the above-mentioned sliding film control rate is used, the force required to grasp the object is substituted into the sliding film control rate to calculate the air pressure, and then the PWM value corresponding to the air pressure is sent to the air pump through the lower computer main control board and the soft artificial hand grasps the object.

[0087] Step (5.3): When the PVDF piezoelectric film sensor at the end of the soft artificial hand detects that the object has a tendency to slide with the soft artificial hand, the holding force of the soft artificial hand end and the object is increased, and step (5.2) is executed until the soft artificial hand completes the task of grasping the object.

[0088] Step (5.4): When the PVDF piezoelectric film sensor at the end of the soft artificial hand detects that the data of the object and the soft artificial hand is less than the sliding threshold value, that is, the soft artificial hand and the object have no tendency to slide, the object grasping is successful. The sliding sensation is the sliding force generated between the object and the soft artificial hand due to sliding, and the sliding threshold value is the value at which the object is about to slide off the soft artificial hand, which is obtained by experiment.

[0089] Compared with the prior art, the beneficial effects of the present application are:

[0090] 1. The mechanical modeling of the pneumatic soft actuator is carried out, the functional relationship of the air pressure and the object contact force of the soft actuator is obtained, and the theoretical basis is provided for the control of the soft prosthetic hand.

[0091] 2. The soft prosthetic hand control system has faster response speed and more stable running state. Since the soft prosthetic hand is made of soft silicone and driven by air pressure, it is a nonlinear system, and at present, no literature can accurately model the soft prosthetic hand. Larger modeling error requires larger switching gain, which causes system chattering. In addition, the traditional PID control algorithm is not suitable for controlling nonlinear systems, and often shows poor controller parameter setting, low control precision, and cannot achieve ideal control effect. The present application proposes a non-singular terminal sliding mode control algorithm combined with RBF neural network algorithm. Firstly, the non-singular terminal sliding mode control algorithm is a nonlinear control algorithm, which is suitable for the control system of the soft prosthetic hand. Secondly, the RBF neural network is used to approximate the unknown function of the modeling error of the soft actuator, which reduces the model disturbance problem, significantly reduces the system chattering, realizes global fast convergence, and makes the control of the soft prosthetic hand faster, more stable and more accurate.

[0092] 3. The soft prosthetic hand control method based on the myoelectric signal is a complete closed-loop control method for controlling the soft prosthetic hand to grasp objects, which is easy to implement, has strong applicability, and is conducive to popularization and use. DETAILED DESCRIPTION

[0093] Figure 1 is a single soft actuator schematic diagram of the present application;

[0094] Figure 2 is a soft prosthetic hand control circuit diagram;

[0095] Figure 3 is a soft actuator bending deformation schematic diagram;

[0096] Figure 4 is a soft actuator bending moment analysis diagram;

[0097] Figure 5 is a soft prosthetic hand force control principle diagram;

[0098] Figure 6 is a non-singular terminal sliding film control algorithm flow chart of RBF neural network approximation disturbance function;

[0099] Figure 7 is a soft prosthetic hand object grasping control flow chart. DETAILED DESCRIPTION

[0100] The present application will be further described below in combination with the drawings and specific embodiments.

[0101] AsFigure 1 The soft actuator shown includes a strain layer, a strain limiting layer, an air channel and an air bag. When the soft actuator is inflated, the strain layer will be deformed, the strain limiting layer will not be deformed, and the soft actuator will bend towards the strain limiting layer side. The soft artificial hand shown is composed of five individual soft actuators, a palm and an arm. The soft actuator mold is made by 3D printing, and then the soft actuator is made by pouring human silicone and embedding sensors. Finally, the five soft actuators are assembled with the palm and the arm to complete the production of the soft artificial hand. Figure 2 The soft artificial hand control system shown includes a lithium battery, an air pump, a solenoid valve, a motor drive chip, a single-chip microcomputer main control board, an air pressure sensor, a thin film pressure sensor, a PVDF piezoelectric film sensor, an electromyographic sensor and a soft artificial hand. The single-chip microcomputer main control board controls the motor drive chip, the air pump and the solenoid valve to inflate and deflate the soft artificial hand to achieve bending and thus grasp objects. The air pressure sensor, the thin film pressure sensor and the PVDF piezoelectric film sensor are embedded in a single soft actuator and are used to collect real-time air pressure in the cavity, contact force between the soft artificial hand and the object and tactile information between the soft artificial hand and the object, respectively. The electromyographic sensor is used to collect electromyographic signals during hand movement. The single-chip microcomputer, the air pressure sensor, the thin film pressure sensor and the PVDF piezoelectric film sensor are all powered by 5V, and the lithium battery powers the entire system.

[0102] As shown in the figure, the dimensions of the soft actuator during inflation and bending are shown in the figure, and the relationship between the stress and the principal elongation ratio of the soft actuator is constructed by the following steps: Figure 3

[0103] Step (1.1): Yeoh model is selected for nonlinear mechanical analysis of the soft actuator. The constitutive relationship of the rubber material is established based on the stress-strain relationship, and the strain energy density function and the deformation tensor invariant of the soft actuator are:

[0104]

[0105] Where W is the strain energy density function, λ1, λ2, λ3 are the principal elongation ratios of the soft actuator in the length, width and height directions, respectively, and I1, I2, I3 are the deformation tensor invariants of the soft actuator in the length, width and height directions, respectively.

[0106] Step (1.2): The principal elongation ratio λ1 of the soft base after the soft actuator is inflated and bent is:

[0107]

[0108] Where l is the length of the air cavity of the soft actuator, t3 is the thickness of the bottom layer of the soft actuator, is the bending angle.

[0109] ​Step (1.3): Combining formula (1)~(2)

[0110]

[0111] λ3=1 (4)

[0112]

[0113] Step (1.4): According to the typical binomial parameter form of Yeoh model, the strain energy density function model of the soft actuator when bending is driven by the software:

[0114] W=C1(I1-3)+C2(I1-3) 2 (6)

[0115] Where C1, C2 are the slope and intercept of the stress fitting straight line of the silicone Yeoh model material, respectively, which are determined by material testing and data fitting. Through experiments, C1 is 0.1 and C2 is 0.005.

[0116] Step (1.5): Combining formula (1)~(6), the stress in the length direction of the soft actuator is:

[0117]

[0118] Combining formula (3)~(7), the relationship between the stress of the soft actuator and the main elongation ratio is obtained:

[0119]

[0120] As shown in Figure 4 , the mechanical analysis of the soft actuator after inflation bending is carried out, and the relationship between the air pressure and the object contact force of the soft actuator is constructed, including the following steps:

[0121] Step (2.1): Constructing the moment equation of the soft actuator. The bending moment of the soft actuator is represented by five bending moments, i.e. the main moment M a generated by the gas acting in the air chamber, the moments M t , M b , M c generated by the stretching of the top layer, bottom layer and side wall materials, and the moment M f generated by the interaction of the end of the soft actuator and the object:

[0122]

[0123]

[0124]

[0125]

[0126]

[0127] where p is the air pressure on the soft actuator air cavity unit, B is the soft actuator width, t is the soft actuator wall thickness, h is the air cavity height, L f is the soft actuator end length, and f is the force generated by the interaction of the soft actuator end and the object.

[0128] Step (2.2): According to the moment balance:

[0129] M a -2M c -M t -M b = M f (14)

[0130] Step (2.3): Combining equations (9)-(14), the soft actuator mechanical equation is described by the following equation:

[0131]

[0132] where g(σ) is the resultant moment generated by the tensile of the top layer, bottom layer, and side wall material of a single soft actuator, f is the force after the soft actuator chamber is inflated, d(t) is the modeling error and unknown function of external disturbance, σ is the stress in the length direction of the soft actuator, L f is the soft actuator end length, and p is the air pressure on the soft air cavity.

[0133] Further, the relationship equation between the air pressure on the soft actuator and the contact force of the object is defined as:

[0134]

[0135] where, define

[0136]

[0137] where is the second-order derivative of the force generated by the interaction of the soft actuator end and the object, is the first-order derivative of the force generated by the interaction of the soft actuator end and the object.

[0138] As Figure 5 , 6 shown, the steps of constructing the non-singular terminal sliding mode function of the deviation of the desired contact force and the actual contact force at the end of the soft actuator are as follows:

[0139] Step (3.1): According to the relationship equation between the air pressure on the soft actuator and the contact force of the object, the control system deviation function is constructed, assuming that the desired force of the soft actuator after being inflated and bent to contact the object is fd The actual feedback force obtained by the thin film pressure sensor is f, and the system bias function is defined as:

[0140] e = f d -f (18)

[0141] Step (3.2): According to the bias function, the sliding mode surface function of the non-singular terminal sliding mode control is selected as:

[0142]

[0143] wherein is a positive number, p and q are positive odd numbers, and Combining equations (16) to (19):

[0144]

[0145] wherein

[0146]

[0147]

[0148]

[0149] As shown in Figure 6 , the steps of the non-singular terminal sliding mode algorithm for approximating the disturbance error function based on the software driver RBF neural network are as follows:

[0150] Step (4.1): The unknown disturbance function is approximated by RBF neural network defined as

[0151]

[0152]

[0153]

[0154] wherein x is the network input, j is the jth node of the network hidden layer, c j is the kernel function center, is the square of the Gaussian kernel width, h(x) = [h j ] T is the network Gaussian kernel output, W * is the ideal weight value of the network, is the estimated value of the ideal weight value of the network, is the error between the ideal weight value and the estimated value of the network, ε is the network approximation error, is the disturbance function approximated by the neural network.

[0155] Step (4.2): Construct the Lyapunov function of the control system as follows:

[0156]

[0157] where γ is a positive number,

[0158] Step (4.3): Calculate the control rate and adaptive rate of the control system when the derivative of the Lyapunov function is less than or equal to zero, that is, When s = 0, it is easy to deduce that the soft actuator control system will continuously approach zero near s = 0, and the derivative of the Lyapunov function is:

[0159]

[0160] Design the control rate

[0161]

[0162] That is,

[0163]

[0164] where sgn() is the sign function, η is the switching term gain, and p is the air pressure acting on the soft actuator cavity.

[0165] Combining the above equations (28)-(30)

[0166]

[0167] Step (4.4): In order to make the non-singular terminal sliding mode controller satisfy the Lyapunov stability condition, that is, Take The adaptive rate is:

[0168]

[0169] Combining equations (31)-(32) gives:

[0170]

[0171] According to the above equation, the controller satisfies the Lyapunov stability condition.

[0172] As shown in Figure 7 The control flow steps of the soft prosthesis hand based on the myoelectric signal to grasp objects are as follows:

[0173] Step (5.1): Use the myoelectric signal acquisition device to collect human myoelectric signals.

[0174] Step (5.2): Preprocessing and action recognition of the collected electromyography signals.

[0175] Step (5.3): The recognized corresponding hand action motion result is taken as a motion instruction, and the instruction is sent to the lower computer main control board. After the main control board receives the instruction, the air pump and the electromagnetic valve control the soft artificial hand to inflate to realize bending and approaching the object.

[0176] Step (5.4): Detecting tactile information. When the data of the film pressure sensor at the end of the soft artificial hand is less than the tactile threshold value, it is considered that the soft artificial hand does not contact the object, and the soft artificial hand continues to inflate and bend. When the data is greater than the tactile threshold value, it is considered that the soft artificial hand is grasping the object. The tactile is the force generated by the contact between the object and the soft artificial hand, and the tactile threshold value is the value at which the soft artificial hand is about to contact the object, which is obtained by experiment.

[0177] Step (5.5): When the soft artificial hand contacts the object, the required grasping force is substituted into formula (30) to calculate the size of the air pressure by using the above synovial membrane control rate, and then the PWM value corresponding to the air pressure is sent to the lower computer main control board.

[0178] Step (5.6): At this time, the air pressure sensor is used to detect the current soft actuator cavity air pressure. When the cavity air pressure is greater than the cavity air pressure limit value, the grasping fails. Otherwise, the main control board calculates the required air pressure PWM instruction to control the air pump to inflate the soft artificial hand to realize the grasping task of the object.

[0179] Step (5.7): When the data of the PVDF piezoelectric film sensor at the end of the soft artificial hand detects that the object and the soft artificial hand are greater than the sliding threshold value, the expected holding force between the end of the soft artificial hand and the object is increased, and step (5.6) is executed until the soft artificial hand completes the task of grasping the object. The sliding force is the sliding force generated between the object and the soft artificial hand due to sliding, and the sliding threshold value is the value at which the object is about to slip off the soft artificial hand, which is obtained by experiment.

[0180] Step (5.8): When the data of the PVDF piezoelectric film sensor at the end of the soft artificial hand detects that the object and the soft artificial hand are less than the sliding threshold value, i.e. the soft artificial hand and the object have no sliding trend, the object grasping is successful.

[0181] In addition to the above embodiments, the present application can have other implementation manners, and any technical solutions formed by equivalent replacement or equivalent transformation shall fall within the protection scope of the present application.

Claims

1. A method of soft prosthesis hand control based on myoelectric signals, characterized by, The method comprises the following steps: (1) collecting the electromyographic signals of the hand movement process through the electromyographic sensor, and identifying the corresponding hand movement results; (2) performing mechanical modeling on the soft actuator, performing nonlinear mechanical analysis on the single soft actuator by using a Yeoh model, and obtaining the relationship between the air pressure and the object contact force borne by the soft actuator according to mechanical equilibrium, wherein the Yeoh model is a constitutive model; (3) constructing a non-singular terminal sliding mode function of the deviation between the expected contact force and the actual contact force at the end of the soft actuator according to the relationship between the air pressure and the object contact force borne by the soft actuator obtained in step (2), so as to accelerate the convergence speed of the control system; (4) for the unknown functions of the control system disturbance and the modeling error, the RBF neural network is used to approximate the unknown functions of the control system disturbance and the modeling error, so as to reduce the chattering problem of the control system, and finally a Lyapunov function of the deviation between the expected contact force and the actual contact force at the end of the soft actuator is constructed, and the control rate and the adaptive rate when the derivative of the Lyapunov function is less than or equal to zero are calculated, thereby adjusting the ideal weight of the neural network according to the adaptive rate and controlling the soft actuator according to the control rate; (5) combining the control rate with the movement identified in step (1) to control the soft prosthetic hand to complete the grasping task.

2. A myoelectrically-based soft prosthic hand control method according to claim 1, characterized in that, In step (2), the mechanical modeling of the soft actuator comprises the following steps: (2.1) selecting the Yeoh model to perform nonlinear mechanical analysis on the soft actuator, establishing the constitutive relationship of the rubber material based on the stress-strain relationship, and then the strain energy density function and the deformation tensor invariant of the soft actuator are: Wherein W is the strain energy density function, λ1, λ2, λ3 are the principal elongation ratios of the length, width and height of the soft actuator respectively, I1, I2, I3 are the deformation tensor invariants of the length, width and height of the soft actuator respectively; The principal elongation ratio λ1 of the soft actuator after the soft base body is inflated and bent is: wherein l is the air chamber length of the soft actuator, t3 is the thickness of the bottom layer of the soft actuator, is the bending angle; Combining formulas (1)-(2) gives: λ3=1 (4) (2.2) According to the typical binomial parameter form of the Yeoh model, the strain energy density function model is: W = C1(I1-3) + C2(I1-3) 2 (6) Wherein C1 and C2 are the slope and intercept of the stress fitting straight line of the silicone Yeoh model material, which are determined by material test and data fitting, and through experiments, C1 is 0.1 and C2 is 0.005; (2.3) The stress in the length direction of the soft actuator is: Wherein σ is the stress, and the stress of the soft actuator is: (2.4) The bending moments of the soft actuator are represented by five bending moments, i.e. the main moment M a , the moments M t , M b , M c , and the moment M f , which are generated by the interaction of the gas in the cavity with the object. where p is the air pressure experienced by the air chamber unit of the soft actuator, B is the width of the soft actuator, t is the wall thickness of the soft actuator, h is the height of the air chamber, L f is the length of the end of the soft actuator, f is the force generated by the interaction of the end of the soft actuator with the object, and x is the increase in wall thickness of the soft actuator. According to the moment balance, we have: M a -2M c -M t -M b = M f (14) (2.5) Further, the mechanical equation of the soft actuator is described by the following equation: where g(σ) is the resultant moment generated by the stretching of the top layer, bottom layer, and sidewall material of the individual soft actuator, f is the force generated by the interaction of the end of the soft actuator with the object, d(t) is the unknown function of modeling error and external disturbance, L f is the length of the end of the soft actuator. Rewrite formula (15) as follows: Wherein, define Further, the relationship equation between the air pressure and the object contact force borne by the soft actuator is defined as: wherein is the second derivative of the force generated by the soft body actuator end interacting with the object, is the first derivative of the force generated by the soft body actuator end interacting with the object.

3. A myoelectrically based soft prosthic hand control method according to claim 1, characterized in that In step (3), the non-singular terminal sliding mode function of the deviation between the expected contact force and the actual contact force at the end of the soft actuator comprises the following steps: (3.1) Constructing the control system bias function based on the relationship equation between the air pressure of the soft actuator and the contact force of the object. Assuming that the expected force of the soft actuator after being inflated and bent to contact the object is f d , the actual force is f, and the system bias function is defined as: e = f d - f (19) (3.2) Selecting the non-singular terminal sliding film function as: wherein is positive, p, q are positive odd numbers, and in combination with equations (18) to (20): Wherein define wherein is an unknown function that needs to be approximated.

4. A myoelectrically based soft prosthic hand control method according to claim 1, characterized in that , the step (4) includes the following steps: (4.1) Approximation of unknown disturbance function using RBF neural network Definitions where x is the network input, j is the jth node in the network hidden layer, c j is the kernel function center, is the square of the Gaussian kernel width, h(x) = [h j ] T is the network Gaussian kernel output, W * is the network ideal weight, is the estimate of the network ideal weight, is the error of the network ideal weight and estimate, ε is the network approximation error, is the interference function approximated by the neural network; (4.2) constructing the Lyapunov function of the control system as: wherein γ is a positive number, (4.3) Design of the sliding mode control law That is Where sgn() is a sign function, η is a switching term gain, and p is the air pressure acting on the soft actuator cavity; Combining equations (21) to (28) (4.4) To make the nonsingular terminal sliding mode controller satisfy the Lyapunov stability condition, take The adaptive rate is: Combining equations (30) to (31), we have: According to the above equation, the controller satisfies the Lyapunov stability condition.

5. A myoelectrically based soft prosthic hand control method according to claim 4, characterized in that , the step (5) includes the following steps: (5.1) taking the corresponding hand motion result identified in step 1 as a motion instruction, sending the instruction to the lower computer main control board, and controlling the air pump and electromagnetic valve to inflate the soft prosthetic hand to achieve bending and approach the object after the main control board receives the instruction; (5.2) when the thin film pressure sensor data at the end of the soft prosthetic hand is greater than the threshold value, it is considered that the soft prosthetic hand is grasping the object, at which point the sliding mode control rate is used to calculate the air pressure by substituting the grasping force of the object into equation (29), and then sending the PWM value corresponding to the air pressure to the air pump through the lower computer main control board to control the soft prosthetic hand to grasp the object; (5.3) when the PVDF piezoelectric film sensor at the end of the soft prosthetic hand detects that the data of the object and the soft prosthetic hand is greater than the sliding threshold, the expected holding force of the soft prosthetic hand end and the object is increased, and step (5.2) is executed until the soft prosthetic hand completes the task of grasping the object, the sliding sensation is the sliding force generated between the object and the soft prosthetic hand due to sliding, and the sliding sensation threshold is the value at which the object is about to slip off the soft prosthetic hand, obtained by experiment; (5.4) when the PVDF piezoelectric film sensor at the end of the soft prosthetic hand detects that the data of the object and the soft prosthetic hand is less than the sliding threshold, i.e., the soft prosthetic hand and the object have no tendency to slide, the object grasping is successful.

Images